Mortality rates over time in published pediatric observational cohorts, predominantly in the Western hemisphere, with sample sizes ≥100 subjects and with similar PARDS-type inclusion criteria. The solid regression line represents a quadratic function showing decreasing mortality over time, with leveling-off in more recent years. The dashed lines represent 95% confidence intervals. The assigned year represents the final year of patient accrual for a given study. The size of the circles represents the relative sample sizes
First, predictors of mortality in PARDS are not necessarily specific to PARDS but, rather, are characteristic of risk factors in several critically ill syndromes. Notably, immunocompromised status [6, 22, 27, 34] and multisystem organ failure (MSOF) [6, 13, 18, 34] are associated with increased mortality risk in several PARDS studies, including RCTs [27]. However, immunocompromised status and MSOF have little pulmonary specificity, are associated with mortality in sepsis, and are components of severity of illness scoring systems. Thus, a generalization of this observation states that children die with PARDS, rather than because of PARDS. In a two-center North American study examining how and why children with PARDS die, neurologic failure (39%) and MSOF (41%) were responsible for the majority of deaths, whereas a minority (20%) died from persistent hypoxemia from refractory PARDS [35]. In many cases, the associated PARDS has resolved at the time of death, despite the persistence of mechanical ventilation. Thus, even the nominally simple notion of “mortality” actually encompasses at least two mutually exclusive competing events: mortality due to PARDS and mortality not due to PARDS, which complicates any inferences made from studies that investigate and report associations between all-cause mortality and an intervention.
Second, elective withdrawal of potentially futile care complicates use of mortality as an endpoint. Withdrawal of care can occur for MSOF, underlying malignancy, or for poor neurologic prognosis, none of which are specific for PARDS. Elective withdrawal was the most common mechanism responsible for death (66%), irrespective of whether the cause of death was neurologic, MSOF, or hypoxemia [35]. While single-center studies may have similar approaches to withdrawal of care, this is more difficult to extrapolate to multicenter or multinational studies, where customs and practices surrounding withdrawal or withholding of care may differ.
One example is worth examining in further detail. A multicenter RCT of exogenous calfactant (bovine surfactant) in moderate and severe PARDS (OI > 7) demonstrated improved mortality associated with calfactant treatment [27]. However, imbalance in the proportion of immunocompromised patients, with overrepresentation in the placebo arm, likely contributed to this effect, and after adjustment for immunocompromised status, the association between calfactant treatment and improved mortality was no longer evident (p = 0.07). Furthermore, patients received treatment within 48 hours of intubation, but the proportion of patients successfully extubated did not differ between the groups, and curves for cumulative successful extubation did not begin to diverge until 12 days after intubation, suggesting that factors unrelated to the initial PARDS insult, such as immunocompromised status [36], may have been responsible for mortality and prolonged ventilation. A follow-up trial of calfactant was restricted to immunocompromised children (CALIPSO: Calfactant for Acute Lung Injury in Pediatric Stem Cell Transplant and Oncology Patients), using mortality as the primary outcome, and was recently stopped for futility due to slow enrollment [37].
While mortality may be problematic as a primary endpoint for a general PARDS population, there are subgroups of children with PARDS who still have a substantial mortality risk, yet with a reasonable chance of survival. CALIPSO was an example of prognostic enrichment: restricting enrollment for a study to a subgroup with a higher predicted severity of illness and more frequent occurrence of the outcome (mortality), thus improving the power to detect an effect of an intervention. CALIPSO limited their intervention to a subgroup of PARDS with high mortality (>50%), albeit at the risk of difficult recruitment and reduced generalizability. Successful trials in adult ARDS of neuromuscular blockade [38] and prone positioning [39] employed this strategy, as ACURASYS (ARDS et Curarisation Systematique) limited enrollment to patients with Pao2/Fio2 ≤ 150, rather than the typical ≤300. PROSEVA (Prone Position in Severe ARDS) required even more stringent enrollment criteria, as it require Pao2/Fio2 ≤ 150 after 12–24 hours of initial stabilization, thereby excluding patients who rapidly improved with standard ventilator management. In both cases, the goal was prognostic enrichment of a higher risk population in which the tested intervention could plausibly impact mortality with a reasonable sample size. This simultaneously avoids unnecessarily exposing patients to treatment when they have low risk of mortality and high probability of survival irrespective of randomization arm, thereby diluting any potential treatment effect. For PARDS to reproduce this, predictors of mortality risk need to be identified and validated. These predictors need to be available early in the PARDS course to allow enrollment within a timeframe amenable for interventions to work, ideally within 48 hours of PARDS onset. This strategy has particular appeal for testing interventions for “refractory” PARDS, such as high-frequency oscillatory ventilation (HFOV), prone positioning, methylprednisolone, inhaled nitric oxide (iNO), and extracorporeal membrane oxygenation (ECMO).
Finally, it is worth discussing why mortality is decreasing in PARDS despite an absence of positive trials. Indirect evidence suggests adoption of management extrapolated from adult ARDS, such as lower tidal volumes [40] and higher positive end-expiratory pressures [41], may be associated with lower mortality. Additionally, as many subjects with PARDS die of MSOF, rather than hypoxemia, it is possible that other temporal changes unrelated to ventilator management have impacted survival, such as protocolized sepsis care and timely antibiotics [42–44]. However, it is also important to note that definitions of ARDS (and PARDS) have evolved over time. The AECC definition [2] allowed for an entity of acute lung injury (Pao2/Fio2 ≤ 300), in addition to ARDS (Pao2/Fio2 ≤ 200), thereby introducing a category of subjects with less severe lung injury. The 2012 revised Berlin definition [3] recoded this category as “mild ARDS,” (200 < Pao2/Fio2 ≤ 300), and introduced the requirement for minimal invasive or noninvasive end-expiratory pressure ≥5 cmH2O. The 2015 PALICC definition of PARDS [4] further liberalized the definition by allowing inclusion of subjects with unilateral infiltrates on chest radiograph, in addition to bilateral. Effectively, the operational definitions of ARDS (and now including PARDS) after 1990 have allowed inclusion of less severe subjects, which may be contributing to the lower mortality rates. Thus, it is entirely possible that the mortality rate for “real” PARDS has not fallen nearly as dramatically as the literature would suggest.
Duration of Mechanical Ventilation
Duration of ventilation is a commonly described outcome in PARDS studies, especially when this outcome is limited to survivors. This outcome has face validity, as more severe PARDS can reasonably be expected to require a longer duration of mechanical ventilation. The 2012 Berlin definition [3] demonstrated an increase in duration of mechanical ventilation in survivors across increasing severity classes of ARDS, which was confirmed in LUNG SAFE (Large Observational Study to Understand the Global Impact of Severe Acute Respiratory Failure) [45]. This observation has been corroborated in PARDS when using oxygenation between 6 and 24 hours, rather than at PARDS onset [46, 47].
To be valid as an endpoint, “duration of mechanical ventilation” needs to be limited to survivors, given the risk of contamination of this outcome with nonsurvivors with a short-duration of ventilation. Furthermore, given the increased utilization of noninvasive ventilation both prior to [48–50] and after endotracheal intubation, duration of mechanical ventilation requires clear definition regarding whether noninvasive support is included. Both Berlin (mild) ARDS [3] and PALICC PARDS [4] definitions make allowances for noninvasive support, suggesting that screening for studies based on these criteria would allow for inclusion of a substantial number of nonintubated patients, some of whom will subsequently be intubated. This potential for increased enrollment needs consistent and well-delineated reporting of what is meant by “duration of mechanical ventilation.”
Therefore, while the endpoint “duration of ventilation in survivors” has face validity and reflects PARDS severity, it is unclear exactly how “patient-centric” this outcome is. Specifically, it is unclear whether a given child would be better served with 10 days of invasive mechanical ventilation and extubated to high-flow cannula, or whether 8 days of invasive ventilation followed by 4 days of noninvasive bilevel positive airway pressure (BiPAP) with full-facemask interface. Indeed, the answer likely varies between patients for a multitude of variables, including sedation requirements, strength, airway status, and indication for intubation.
Finally, duration of ventilation in survivors is complicated by the prevalence of subglottic stenosis, poor secretion tolerance, or severe upper airway obstruction from poor airway tone as an indication for prolonged intubation. Such patients may wean appropriately to minimal invasive support given their underlying PARDS severity, but the actual act of removing the endotracheal tube may be delayed, or ultimately attempted and unsuccessful, for reasons related primarily to their airway. Given the substantial number of comorbidities described in PARDS [46], reasons for prolonged intubation unrelated to the actual PARDS risk factor have the potential to confound the utility of duration of ventilation as an endpoint. An alternative has been proposed to only count the duration of time until successful completion of an extubation readiness test, irrespective of whether or not the patient is actually extubated [51]. However, this has not been validated nor described in an actual practice or trial, and does not address the prior criticism of not being patient-centered, as the child remains intubated.
Ventilator-Free Days
One of the most commonly adopted composite endpoint in PARDS trials is ventilator-free days (VFDs), typically at 28 days. VFDs at 28 days are derived by subtracting ventilator duration in survivors from 28, and scoring nonsurvivors and those requiring ≥28 ventilator days as 0 [52]. It has also been defined as “days alive and free of mechanical ventilation” [53], which creates confusion for cases where the patient is extubated on day 10, but dies on day 20 (10 days alive and free of mechanical ventilation is VFD = 10; nonsurvival at day 28 suggests VFD = 0). This endpoint combines mortality and duration of ventilation by penalizing nonsurvivors, unlike duration of ventilation. Similar to duration of ventilation in survivors, VFDs at 28 days have demonstrated correlation across severity of Berlin ARDS [3] and PALICC PARDS [46, 47] categories, with worse oxygenation categories associated with fewer VFDs. This composite endpoint demosntrates efficiency, as an outcome of an intervention which both reduces mortality and duration of ventilation can be detected with a smaller sample size [53].
The same caveats regarding clarity of noninvasive support are required for VFDs as mentioned for duration of ventilation in survivors [52]. However, VFDs has a major limitation as a composite endpoint, as the merged individual endpoints (mortality and ventilator duration) are not equivalent and interchangeable. A child requiring 30 days of mechanical ventilation, but surviving, cannot be considered identical to a child who dies after 7 days of ventilation, although both would be recorded as VFD = 0. Composite endpoints are best utilized when the separate endpoints are of equivalent importance for the patient, such as stroke or myocardial infarction in hypertensive adults. When initially described for adult ARDS, VFDs were demonstrated to be useful only when the more pejorative outcome of mortality was improved alongside shorter duration of ventilation [53]. Given the >30% mortality in adult ARDS [45], this is a reasonable expectation: interventions, which shorten ventilation should improve mortality, assuming mechanical ventilation and ARDS are in the causal pathway for nonsurvival. However, even in adults, this assumption can be problematic. The ARDSNet corticosteroid trial [54] failed to demonstrate superiority of methylprednisolone for persistent ARDS for the primary outcome of mortality at 60 days (29.2% mortality in methylprednisolone, 28.6% in placebo, p = 1). However, methylprednisolone was associated with 4.4 additional VFDs and 2.7 additional ICU-free days at 28 days. Significantly more patients in the methylprednisolone arm required re-initiation of ventilation (28% vs. 9%, p = 0.006). These discrepant results make interpretation of the trial difficult: mortality is reported at 60 days, but VFDs at 28 days. Mortality is nominally higher in the methylprednisolone group, but VFDs are also more favorable for methylprednisolone. Thus, in this case, the reporting of VFDs offers no advantages or power relative to reporting on mortality alone: when an intervention has opposite effects on duration of ventilation and mortality, VFDs merely confuse the interpretation.
In pediatrics, the use of VFDs is potentially suspect for these same reasons, as PARDS mortality is much lower, and persistent hypoxemia is unlikely to be the cause of mortality [35]. Thus, the effect on mortality is less certain to be in the same direction as duration of ventilation. For instance, a trial of ECMO for severe refractory PARDS may result in nominally improved mortality rates but would likely result in prolonged duration of ventilation, thereby complicating the interpretation and utility of VFDs. Finally, several interventions sorely in need of testing in PARDS, including fluid management, sedation protocols, weaning, and extubation readiness, all clearly impact length of ventilation much more so than they will impact mortality, hampering the utility of VFDs as an outcome unless these parameters are protocolized in the context of the trial.
Analysis of VFDs is also not straightforward. VFDs are typically analyzed by comparing means or medians, using t-tests or rank-sum tests, respectively. The skew, excess of zeroes, and ordinal nature of VFDs complicate the use of parametric tests, like t-tests, whereas the nonparametric equivalents, like the rank-sum tests, do not readily allow for covariate adjustment or efficient description of effect size. Alternative approaches, such as competing risk regression, in which successful extubation is the primary outcome, and death is treated as a competing event, may overcome some of the limitations of traditional tests [55]. Analyzing VFDs in a competing risk framework treats extubation as a time-to-event analysis, censoring after day 28, with nonsurvivors set to be “never extubated” sometime after day 28. This is parallel to setting nonsurvivors to VFD = 0. This framework is less affected by skew or zero-inflation, readily incorporates additional covariates, and clearly imparts information regarding effect size.
Need for Extracorporeal Support as an Outcome
Another composite outcome for PARDS investigations has been the combination of need for ECMO or death [19, 29]. This attempts to address the limitations of VFD and the low mortality (and thus difficult to adequately power) of PARDS. The underlying assumption is that lung injury severe enough to require ECMO is essentially refractory to conventional mechanical ventilation, and thus, need for ECMO would be a death in any center unable to provide ECMO. Therefore, “ECMO” is close enough to “death” to justify combination as a composite endpoint.
The European Society for Pediatric and Neonatal Intensive Care used this definition to test the utility of the Berlin criteria in children [19], and demonstrated that the inclusion of a “severe” ARDS category improved validity with an increased risk of ECMO/death in children with Berlin-defined severe ARDS. It should be noted, however, that the incidence of ECMO/death (18.6%) was only marginally increased over the incidence of mortality (17.2%), and that comparable analyses for mortality yielded identical conclusions.
A recently published RCT [29] for iNO (total n = 53) reported both mortality (28% placebo, 8% iNO, χ2 p = 0.07) and ECMO/death (48% placebo, 8% iNO, p < 0.01). The trial was powered for a difference in VFD at 28 days, for which it required a sample size of 169 children, and was stopped early for slow enrollment. Of note, the difference in the reported VFD in this trial was also significant. While the primary outcome of more VFD was achieved despite the small sample size, the reporting of ECMO/death in this study points to a potential mechanism whereby iNO improved VFD. Specifically, iNO appeared to decrease the rate of ECMO utilization, suggesting an improvement in hypoxemia, thereby reducing total ventilator days, and potentially impacting mortality. This is significant, as it implies a connection between improvement in hypoxemia and better outcomes in PARDS, a connection that is not consistently observed in adult ARDS trials [56]. The recently completed ECMO to rescue Lung Injury in severe ARDS (EOLIA) trial in very severe, refractory adult ARDS was stopped early for futility for a low probability of achieving its primary endpoint, mortality at 60 days, despite a nominal improvement in mortality with ECMO (relative risk with ECMO 0.76, 95% confidence interval [CI] 0.55 to 1.04, p = 0.09). However, 28% of subjects assigned to mechanical ventilation crossed over to ECMO, and when the trial was reanalyzed using “treatment failure” as the outcome, the result was highly significant in favor of ECMO (relative risk 0.62, 95% CI 0.47 to 0.82, p < 0.001). The authors defined treatment failure as death for the ECMO arm, and as death/ECMO for the ventilation arm, providing face validity for this outcome in future trials.
For certain trials of salvage therapy, such as methylprednisolone, iNO, prone positioning, and HFOV, the use of ECMO/death as a primary outcome may be rational. However, as in the iNO trial example above, there is little information added by this specific reporting that was not also captured by the more conventional short-term outcome of VFD at 28 days. Additionally, as ECMO is not an outcome per se, but simply an additional mode of supportive care, with subjective thresholds for its utilization among different centers and practitioners, the composite outcome of ECMO/death is difficult to standardize. Finally, the component variables of ECMO/death are not of equal importance to the patient, thus calling into question its validity as a patient-centered, clinically meaningful composite outcome, similar to the criticism of VFDs.
Technology Dependence and New Morbidity as an Outcome
Development of new morbidity, quantified using scoring systems such as the Functional Status Scale (FSS), which penalize additional technology dependence, has been proposed as outcome for trials in critically ill children [57]. FSS scores subjects from 1 (normal function) to 5 (severe dysfunction) points across six domains (mental status, sensory function, communication, motor function, feeding, respiratory). New morbidity, operationalized as an increase in FSS from baseline of 3 or more points from baseline, was demonstrated to occur nearly 1.5- to 2-fold more frequently than mortality.
In a single-center study of 316 subjects with PARDS [58], new morbidity (ΔFSS ≥3 from baseline) occurred in 20% of subjects, whereas hospital mortality occurred in 13%. Thus, use of death and new morbidity as a composite outcome would have nearly tripled the event rate for a trial, from 13% to 33%, demonstrating utility of new morbidity as a viable outcome for PARDS. In this PARDS cohort, worsening in the FSS domains of motor function, feeding, and respiratory was associated with discharge to a location other than home.
A criticism of new morbidity is that it is only partly related to the acute PARDS event, but is also substantially impacted by underlying comorbidities. Of the 274 survivors in this study, 56 (18% of the entire cohort; 20% of survivors) had a worsening respiratory FSS, of whom 19 (6% of the entire cohort; 7% of survivors) underwent new tracheostomy placement. Worsening respiratory FSS, which in practice means increased use of supplemental oxygen or varying degrees of noninvasive and invasive respiratory support, may be more directly related to PARDS, and could potentially be combined with death as part of a composite. The criticism of all composite outcomes that the components are of unequal importance, however, still persists, as tracheostomy is typically not considered equivalent to death.
Postdischarge Outcomes
Potential outcomes for PARDS studies
Outcome | Timeframe | Advantages | Disadvantages |
|---|---|---|---|
Mortality | Short term: 28 or 60 day PICU Hospital | Easy to obtain Fixed time-point Related to acute process Patient-centered | Impractical given low baseline mortality |
Medium and long term: 90 day 1 year | Potentially captures longer period of risk for unfavorable outcomes | Low postdischarge mortality Harder to obtain follow-up More related to underlying comorbidities | |
Ventilator-free days | 28 days | Easy to obtain Increases power to detect clinically meaningful improvements related to shortened ventilation and survival | Needs noninvasive support explicitly defined Imbalance in components of the composite outcome Only increases power if intervention benefits both mortality and ventilator days Specific analytic techniques |
Ventilator days | 28 days PICU LOS | Easy to obtain Related to pulmonary nature of PARDS | Needs noninvasive support explicitly defined Unclear if patient-centered Ignores mortality |
ECMO/death | Short term | Increases power to detect efficacy of pre-ECMO “salvage therapies” | Subjective use of ECMO Imbalance in components of the composite outcome Unclear if patient-centered |
Neurocognitive and functional (POPC/PCPC) | Medium and long term: 90 day 1 year Pre-return to school | Rapid (POCP/PCPC) Patient-centered Potentially completed over telephone Potentially more practical, as it is a prevalent outcome | More thorough cognitive function requires longer testing Changes with developmental age and with comorbidities |
Pulmonary outcomes | Medium and long term: 90 day 1 year Pre-return to school | Patient-centered Related to pulmonary nature of PARDS | Requires infrastructure (expertise and equipment) for in-person follow-up |
Need for tracheostomy | Short, medium, and long term: Hospital discharge 1 year Prereturn to school | Easy to obtain Patient-centered More directly related to PARDS diagnosis High probability of return to clinic for follow-up | Low event rate Collaboration with specialties required |
New morbidity | Short, medium, and long term: Hospital discharge 1 year Prereturn to school | Easy to obtain More frequent than mortality Patient-centered Partly related to pulmonary nature of PARDS | Definitions of morbidity require long-term validation Potentially more related to underlying comorbidities than to PARDS |
Psychiatric | Long term | Patient-centered Potentially completed over telephone | Requires infrastructure (expertise) for in-person follow-up |
Health care utilization | Medium and long term: 90-day and 1-year Rehospitalization | Patient-centered Does not require inpatient follow-up Related to pulmonary nature of PARDS Addresses cost to patient/family | Difficult to obtain Sensitive to local practices Potentially more related to underlying comorbidities than to PARDS |
Stay updated, free articles. Join our Telegram channel
Full access? Get Clinical Tree
